TRACKING SPECIALIST — BARIS ASA · LONDON
Synctag
Start the check → TRACKING SPECIALIST — BARIS ASA · LONDON

READING ROOM › SYMPTOM FILES › SYM-003

SYMPTOM FILE SYM-003 FILED 09 JUN 2026

GA4 withholds your smallest segments: data thresholding.

ALSO FILED AS
  • “ga4 data thresholding applied”
  • “why is ga4 hiding my data”

Diagnosis

GA4 is not losing the data. It is withholding it. When a report could single out an individual, GA4 applies data thresholding: it drops the small rows, and the visible numbers stop summing to the total. It kicks in when a row’s user count is too low to stay anonymous, typically with demographics, search query data, short date ranges, or Blended and Observed reporting identity without enough signed-in activity. The hits are collected in full. The report just refuses to show them at that grain.

This is not sampling, and it is not the (other) row. Sampling is what happens to a query that exceeds the event quota, mostly in Explorations. (other) is a table past its row limit, usually a high-cardinality dimension. Thresholding is a privacy rule on small counts. Same complaint of “missing data,” three causes, three fixes, all filed under R-502 .

SOURCE / MEDIUMBLENDEDDEVICE-BASED
google / organic4,2104,210
newsletter / emailwithheld38
partner-x / affiliatewithheld22
SAME REPORT, TWO REPORTING IDENTITIES · GA4, anonymised

The newsletter and affiliate rows were there the whole time. Under Blended identity, GA4 held them back to protect a handful of users. Under Device-based reporting they come back.

Where it shows

A “thresholding applied” note in the data quality indicator above a report, or named rows that will not add up to the total. The smaller the segment, and the shorter the date range, the more likely it disappears. The tell: the data is fine in aggregate and only goes missing when you drill down.

℞ The fix

  1. Compare the property under Device-based reporting identity. Reporting identity is a property setting: in Admin, set it to Device-based instead of Blended or Observed. Device-based relies on the device ID alone, and Google ties Blended and Observed thresholds to signed-in activity, so small rows often come back. The switch does not change collected data and can be undone. You trade some cross-device detail for rows you can actually see.
  2. Pull the raw data from BigQuery. The GA4 BigQuery export is event-level raw data, with no sampling and no (other) row. It leaves out Google signals data, so expect a small gap to the interface. For any report where the hidden rows matter, query the export instead of the UI.
  3. Widen the range or coarsen the breakdown. Thresholding hits low counts. A longer date range, or one less dimension in the breakdown, often lifts the row back over the threshold.

Thresholding is a privacy feature doing its job, not a bug. The fix is not to defeat it, it is to ask GA4 the question in a form it is allowed to answer. If the totals themselves look wrong, rather than just the breakdowns, that is a different problem, and it is filed under R-501 .

← back to the reading room